1

I share a 12 core Redhat system with a research group. We've recently been running into problems cordially sharing the computational resources, so I'm wondering if it's worth installing a PBS that will schedule jobs and limit individuals' usage. I have some questions about the process of installing the system:

  • Are there any costs associated with running a PBS? For instance, I'll probably have to specify a head node that will no longer be usable for computing (its computing power will be tied up in login management & job distribution) -- are there any other costs that I should think about?
  • Do you have any good resources for correctly installing the PBS? I've clicked around various Google search results but I don't feel like I know what I'm doing. A reliable & easy-to-follow guide would be appreciated.
  • 1
    Other costs include folks forgetting how to use it (e.g. if the IT guy graduates and then nobody knows how to use the scripts), the complexity of the install and maintenance versus solving the problem via some other method (whiteboard, wiki, bribes of beer)... – thrig Nov 10 '16 at 17:38
1

PBS is good but it can take some effort to set up. Try looking into the following:

  • ROCKS clusters - a CentOS-based compute cluster OS; takes the effort out of setting up PBS
  • Torque cluster manager and Maui - ROCKS uses these under the hood; you should be able to install these without switching to ROCKS
  • Installing a new OS seems a bit overkill, but Torque looks like a good alternative. I'll look into it. Why do you say that PBS can take effort to set up? Is it really more effort than switching to a new OS like ROCKS? – alexvas Nov 10 '16 at 20:34
  • IIRC, you have to install the pbs monitor on the head node, then install it individually on all the compute nodes, then you have to tell the head node how to communicate with all the pbs daemons on the cluster nodes, and then if something goes wrong with the daemons on the cluster nodes, you have to manually troubleshoot (in a rocks cluster, you just nuke the node and rebuild it). rocks automates all of that, which saves a lot of effort. – jayhendren Nov 10 '16 at 21:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.